State of the art of satellite rainfall estimation

Transcription

1 State of the art of satellite rainfall estimation 3-year comparison over South America using gauge data, and estimates from IR, TRMM radar and passive microwave Edward J. Zipser University of Utah, USA With acknowledgments to Galdino Viana Mota, Stephen Nesbitt, Chris Kummerow, Wesley Berg, and the entire TRMM science team

2 If you believe that IR estimates are OK, you may take a coffee break now. (Actually, many rainfall estimates are quite accurate if one averages over a sufficiently large spacetime domain)

3

4

5 Outline of talk Compare different algorithms over South America for the 3-year period Dec Nov 2000 Demonstrate that some of the differences are functions of type of precipitation system and type of meteorological regime Summarize a few findings for South America Summarize the (unsatisfactory) state of the art

6 Rain Estimation Issues Global gauge coverage is sparse, especially in the deep tropics

7

8

9

10

11

12 REGIONAL COMPARISONS FOR S.A. (G.V. Mota, 2003) RA IN FA LL ESTIMA TES N ORTH OF 15S mm/mon 50 0 D JF MAM JJA SO N MEAN SEASO N GA UGES GP C C PR TMI GP I RA IN FA LL ESTIMA TES SOUTH OF 15S mm/mo 10 0 n 50 0 DJF MAM JJA S ON S EAS ON GA UGES GP C C PR TMI GP I MEAN

13 SOME LONG-TERM & REGIONAL COMPARISONS TMI and GPI are the highest in the regions of maximum rainfall (higher than the PR and the gauges) GPCC misrepresents rainfall maximum (comparing with the climatologies) in the regions with lack of stations. Good qualitative agreement is found between PR and the Climatologies showing the position of rainfall maxima. PR estimates are a little lower than gauges in the tropics and a little higher than gauges in the subtropics.

14 DJF 3-YR Average Monthly Rain (mm) PR TMI

15 / PR DJF 3-YR AverageTMI GPI / PR

16 There is every reason to believe that the biases between estimation methods are strong functions of the meteorological regime. Therefore, we have chosen to subdivide the precipitation into specific features (PFs), and to classify them according to their properties First step: analyze mesoscale convective systems (MCSs) and compare their properties with smaller and less organized systems

17 PR/TMI Global Difference Map ( from 3G68 Dataset) (courtesy Wesley Berg)

18 The Precipitation Feature Algorithm The PF algorithm was originally designed to synergize the TRMM PR, TMI, and LIS data to identify and classify storms by their size and intensity within the PR swath (Nesbitt et al. 2000) PR and TMI pixels are matched using a nearest neighbor technique, adjusting for parallax The Parallax Problem Original Satellite Geolocation Parallax Adjusted Geolocation Shifted 1 Scan

19 Precipitation Features What is a precipitation feature? Contiguous area at least 4 pixels in size (75 km 2) with: PR near surface reflectivity 20 dbz (identify near surface rain) or TMI 85 GH z PCT 250 K (identify anvils and be consistent with previous work using the SSM/I, e.g. Mohr and Zipser 1996)

20 Classification of Precipitation Features (PFs) PF without Ice Scattering (NI) No pixels with 85 PCT 250 K PF with Ice Scattering (W I) At least 1 pixel with 85 PCT 250 K PF with a Mesoscale Convective System (MCS) M eets ice scattering area and intensity criteria of M ohr and Zipser (1996) Adapted from Nesbitt et al. (2000)

21

22

23

24 Subtropical South America has the largest fractional contribution of PFs with MCSs to rainfall of anywhere on earth between 36 N and 36 S

25 Feature-by-Feature Biases by Feature Type

26 Location of Features with Largest Absolute Differences

27 DJF % Rainfall w/mcs w /LIS

28 Max dbz DJF Median Max dbz at 9 km & Flash Rate in MCSs Flash rate

29 Summary Satellite estimates of monthly rainfall are improving, but still have regional uncertainties and biases of ~ 20-30% Satellite estimates of rainfall from any specific system may still have uncertainties and biases of ~ factor of 2 MCSs dominate rainfall in some areas, and they tend to have a consistent bias in various satellite algorithms, but we must do further work to focus on the reasons The various TRMM algorithms can be used to learn a great deal more about the structure of precipitation systems. For example, there is an important difference in convective intensity between the rain systems in the SACZ compared with those in the Chaco and Argentina

30

31 Attenuation in Surface Clutter Height 30 dbz V5: h1 h2 PIA estimate from SRT V6: 0 Constant slope of Z after correction Surface clutter 0 Stratiform: db/km Convective: 0 db/km

32 Z-R relations (DSD) used in V5 and V6 V5. Conv V5. Strat T. Kozu

33 TMI Tb s TMI 2A12 Ver. 5 Algorithm Ocean: 9 channels/ c-s classification Obs T b Land: NESDIS Operational rainfall-ice scattering relationship derived from radar (Ferraro and Marks 1995) RR= (SI) SI=[ Tb19V Tb22V (Tb22v)2 ] Tb85V CRM T b Bayesian Model 4 Model simulations from Goddard Cumulus Ensemble model CRM profile Surface rain rate LH profile from Kummerow et al. (2001, JAM)

34 The 3-Year PF Database The Ver. 5 database extends from Dec Nov. 2000; Ver. 5A database extends from Mar-Aug 2002 Season Total w/mcs w/ice scattering w/o ice scattering DJF MAM JJA SON Total Stored for each PF are ~30 characteristics: i.e. min 85 GHz PCT, max 30 dbz height, time of occurrence, area, total volumetric rainfall, etc.

35 Typical MCS over Amazonia for which TMI and IR estimate ~40% more rain than TRMM radar (PR)

36 Convective MCS Profiles over Africa and South America

37 Stratiform MCS Profiles over Africa and South America

38 INTRODUCTION: CLIMATOLOGY Rain Gauges => conventional and field experiments - Lack and/or sparseness of rain gauges are obstacles Distribution and characteristics of indirectly estimated precipitation - Outgoing Longwave Radiation (OLR) and/or meteorological analysis to describe large-scale distribution and variability of: - convection, - rainfall, and - MCCs It is well known that the cloud-top temperatures measured remotely by IR do not describe directly the physical processes occurring in clouds and their consequent precipitation.

39

40 Location of Features with Largest Fractional Differences

41 TMI V6 TMI V5 PR V (ITE-97) PR V5

42 Polarization Corrected Temperature To remove non-uniform surface emissivity effects, a Polarization Corrected Temperature is calculated from the 85 GHz horizontally and vertically polarized T b s (Spencer et al. 1989): 85 H 85 V 85 PCT

43 2.5 Estimates

44 2.5 Estimates where Gauges Exist

45 All Season Rainfall Estimates

46 Likely path of overshooting top